A wide-field-of-view camera image reconstruction method based on top-down perspective

Abstract

This invention discloses a wide-field-of-view camera image reconstruction method based on a top-down perspective, belonging to the field of computer vision and image processing technology. The method includes: constructing a top-down imaging geometric model to obtain full and partial line-of-view maps; dividing the imaging region into grids; constructing a weight matrix reflecting the depth and density of the line-of-view; using the existing radiation distribution as initial values ​​and combining it with the weight matrix to generate an initial image; and continuously updating the radiation distribution through an iterative algorithm until the reconstruction error converges. This invention, through line-of-view transformation and iterative solution, not only obtains a wider field of view but also achieves high-precision image reconstruction under complex line-of-view conditions.

Description

Technical Field

[0001] This invention belongs to the field of computer vision and image processing technology, specifically relating to a method for wide-field-of-view camera image reconstruction based on a top-down perspective. Background Technology

[0002] In nuclear fusion experimental devices, especially tokamak research, accurate measurement and reconstruction of the radiation distribution of neutral particles and impurities in the divertor region are crucial. This is essential for a deeper understanding of the density profile of the sump region, the behavior of localized modes at the edges, and the physical state of the scraped layer. Currently, tomographic reconstruction techniques based on optical cameras are widely used for imaging diagnostics of such internal plasma radiation.

[0003] Traditional image reconstruction methods typically rely on tangential or lateral viewing angles. In such geometric configurations, the camera's line of sight is roughly parallel to the main structural surface of the device. While this viewpoint offers good spatial resolution for certain local structures, its inherent limitations are becoming increasingly apparent: First, the effective field of view (FOV) is severely limited, making it difficult to cover large areas in a single image, especially resulting in incomplete acquisition of structural information extending along the circumferential or polar directions; second, in imaging environments with complex three-dimensional structures (such as divertor targets or cooling pipes), the tangential line of sight is easily obstructed by foreground objects, leading to the loss of deep or back-facing radiation information, a problem known as "line of sight occlusion." This directly affects the completeness and accuracy of the reconstructed image.

[0004] To reconstruct two-dimensional or three-dimensional radiation distributions from limited projection data, iterative reconstruction algorithms (such as the maximum likelihood expectation maximization algorithm) are widely used. However, when the observation data itself contains significant information gaps due to viewpoint limitations, the reconstruction quality of such algorithms will deteriorate significantly. Existing methods typically exhibit strong dependence on initial guesses, slow convergence speed, and generally high reconstruction errors (such as mean relative error, MRE) under limited viewpoint conditions, making it difficult to meet the requirements of high-precision physical analysis.

[0005] Therefore, there is an urgent need in this field for a novel imaging method that can fundamentally expand the field of view, reduce information loss, and achieve high-precision reconstruction by combining robust algorithms with hardware improvements. Changing the relative geometric relationship between the camera and the target, thus breaking free from the constraints of traditional tangential perspectives, becomes a potential approach to solving the aforementioned problems. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a wide-field-of-view camera image reconstruction method based on a top-down perspective. By changing the observation geometry, the original tangential perspective is transformed into a top-down perspective, thereby obtaining a wider field of view. Furthermore, it combines an iterative optimization algorithm to improve reconstruction accuracy.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A method for wide-field-of-view camera image reconstruction based on a top-down perspective includes:

[0009] Step 1: Configure the camera from a tangential viewpoint to a top-down viewpoint. By adjusting the position of the optical center and the pitch and azimuth angles of the direction vector, construct the camera imaging geometry model from the top-down viewpoint to obtain full and partial line-of-sight information.

[0010] Step 2: Divide the target's imaging space into a grid, converting the continuous scene into discrete grid units;

[0011] Step 3: Based on the grid division results and the line-of-sight information from the top-down view, construct a weight matrix that reflects the depth and density of the line of sight;

[0012] Step 4: Combine the weight matrix with the input initial radiation distribution vector to generate the initial image vector;

[0013] Step 5: Based on the initial image vector, with the goal of minimizing the difference between the original observed signal and the reconstructed signal, the radiation distribution is continuously updated through an iterative algorithm until the reconstruction error converges, and the final reconstruction result is output.

[0014] Furthermore, obtaining partial line-of-sight information in step 1 includes: truncating the entire line-of-sight by calculating the intersection point between the entire line-of-sight and the boundary of the target area, and obtaining line segments located only within the target area as partial lines of sight; if a certain line-of-sight does not pass through the target area, the pixel corresponding to the line-of-sight is marked as invalid in the reconstruction calculation, and its corresponding row in the weight matrix is ​​all zero.

[0015] Furthermore, step 2, which involves dividing the imaging space into grids, includes: defining a rectangular bounding box on the poloidal section of the target, dividing it into square pixel grids of uniform size, and utilizing the axial symmetry of the plasma to extend each square grid around the circumference, forming a ring with a square cross-section.

[0016] Furthermore, the construction of the weight matrix in step 3 includes: using a fast ray tracing method based on grid topology adjacency, first determining the initial incident surface and incident point of the line of sight entering the modeling region; based on the topological continuity that the exit surface of the current grid is the incident surface of the next adjacent grid, deriving the spatial index of the next grid; calculating the chord length of the line of sight inside each grid as the weight, and obtaining the weight matrix.

[0017] Furthermore, the steps for determining the initial incident surface and incident point include: calculating the intersection points of the ray with the six surfaces of the first grid; if there are two intersection points, calculating the Euclidean distance between the two intersection points and the ray source respectively; the point with the smaller distance is determined as the initial incident point, and the surface where it is located is the initial incident surface; the point with the larger distance is determined as the exit point.

[0018] Furthermore, after constructing the weight matrix in step 3, the method further includes: deleting columns in the weight matrix where all columns are zero, so as to delete grid cells that are not visible to all lines of sight; and simultaneously deleting rows in the weight matrix where all rows are zero, so as to delete lines of sight that do not pass through the target area.

[0019] Furthermore, the iterative algorithm in step 5 employs the conjugate gradient least squares method, including:

[0020] Initialization phase: Set the initial radiation distribution vector, calculate the initial measurement residual based on the weight matrix and the initial radiation distribution vector, calculate the initial projection residual based on the transpose of the weight matrix and the initial measurement residual, and set the initial conjugate search direction vector to be equal to the initial projection residual;

[0021] Iteration phase: In each iteration, the step size factor is calculated based on the relationship between the current projection residual, the weight matrix, and the current search direction vector. The reconstruction result, measurement residual, and projection residual are updated sequentially using this step size factor. The conjugate orthogonalization weight coefficient is calculated based on the norm ratio of the projection residual before and after the update. Then, the updated projection residual is combined with the current search direction vector based on this weight coefficient to construct the next conjugate search direction. The above iteration phase is repeated until the preset iteration termination condition is reached.

[0022] Furthermore, the method for determining the convergence of reconstruction error in step 5 is as follows: calculate the average relative error between the initial image vector and the reconstructed image vector, and determine convergence when the average relative error meets a preset threshold.

[0023] In a second aspect, the present invention provides an electronic device comprising: one or more processors; and a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned wide-field-of-view camera image reconstruction method based on a top-down perspective.

[0024] Thirdly, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enable the processor to implement the aforementioned wide-field camera image reconstruction method based on a top-down perspective.

[0025] The beneficial effects of this invention are as follows:

[0026] Effectively expand the field of view: By switching the camera's perspective from traditional tangential observation to top-down observation, the mutual occlusion of the line of sight in complex scenes is significantly reduced, allowing a single image to cover a wider target area, thereby obtaining more complete and macroscopic scene structure information.

[0027] Significantly Improved Reconstruction Accuracy: The top-down perspective combined with an optimized iterative reconstruction algorithm employed in this invention effectively improves the completeness of the projection data and the conditions of the weight matrix, making the reconstruction process more stable and accurate. Experimental verification shows that this method can significantly reduce the mean relative error (MRE) of the reconstructed image to approximately 0.18%, which is far superior to the reconstruction results under the traditional tangential perspective.

[0028] Enhancing the robustness and applicability of the method: This method does not rely on precise guesses of the prior structure of the scene; it can stably converge to a high-precision solution by iterating from a randomly initialized radiation distribution. This characteristic of not relying on specific initial values ​​makes it highly adaptable and robust to complex, unknown, or dynamically changing radiation fields.

[0029] Providing new solutions for related fields: This method is not only applicable to radiation imaging diagnosis in nuclear fusion devices, but its core idea of ​​"viewpoint transformation + iterative solution" also provides new technical references for image reconstruction tasks in computer vision, industrial inspection and medical imaging fields where there are limited or obstructed fields of view. Attached Figure Description

[0030] Figure 1 This is a flowchart of a wide-field-of-view camera image reconstruction method based on a top-down perspective according to the present invention.

[0031] Figure 2 This is a diagram showing the transformation from a tangential view to a top-down view in an embodiment of the present invention, where (a) is a tangential view and (b) is a top-down view;

[0032] Figure 3 The above are the grid line-of-sight distribution map and the camera top-view line-of-sight depth map in the embodiments of the present invention, wherein (a) is the number of lines of sight traversed by each grid, and (b) is the camera top-view line-of-sight depth map;

[0033] Figure 4 This is a schematic diagram of a three-dimensional model and a mesh on one side of the model established in the embodiment of the present invention, wherein (a) is a three-dimensional model and (b) is a mesh diagram on one side of the three-dimensional model;

[0034] Figure 5 The images show the initial poloidal cross-sectional radiation distribution and the inverted poloidal cross-sectional radiation distribution in this embodiment of the invention, where (a) is the initial poloidal cross-sectional radiation distribution and (b) is the inverted poloidal cross-sectional radiation distribution.

[0035] Figure 6 These are the original camera image and the inverted image in this embodiment of the invention, wherein (a) is the original camera image and (b) is the inverted image;

[0036] Figure 7 This is a graph showing the change of log(MRE) with the number of iterations in an embodiment of the present invention. Detailed Implementation

[0037] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0038] like Figure 1 As shown, this invention provides a method for wide-field-of-view camera image reconstruction based on a top-down perspective, the specific process of which is as follows:

[0039] Step 1: Construct a top-down camera imaging geometry model, configuring the camera from a traditional tangential viewpoint to observe the target from a top-down perspective, such as... Figure 2 As shown, the imaging geometry model is first adjusted, where (a) is the tangential view and (b) is the top-down view. By adjusting the optical center position and the pitch and azimuth angles of the direction vector, the original tangential observation view is converted into a top-down observation view. This reduces occlusion caused by tangential observation and expands the effective field of view. This viewpoint transformation allows the camera to capture a wider range of scenes and reduces mutual occlusion between objects.

[0040] This invention acquires the corresponding full line of sight and the intercepted partial line of sight information of the camera. The "full line of sight" is the camera's theoretical field of view, while the "partial line of sight" is the camera's effective detection path. Since the camera's line of sight extends infinitely in space, while the target area to be reconstructed is finite, this invention calculates the intersection point of the full line of sight with the boundary (or physical constraint wall) of the target area, truncates the full line of sight, and obtains a line segment located only within the target area, i.e., the "partial line of sight."

[0041] When constructing the weight matrix W (in subsequent step 3), these partial views are used for calculation:

[0042] 1. Determine the effective computation domain: If a certain line of sight is not partially captured (i.e. does not pass through the target area), the pixel corresponding to that line of sight is marked as invalid or background noise in the reconstruction calculation, and its corresponding row in matrix W is all zero.

[0043] 2. Depth Information Extraction: The "camera line-of-sight depth map" is as follows: Figure 3 Figure (b) is generated based on the geometric length of a partial line of sight. The longer the partial line of sight, the more meshes it passes through and the deeper it is. Figure 3 As shown in Figure (a).

[0044] 3. Weight calculation: For matrix element W ijIts value depends on the chord length of the p-th line of sight (i.e., part of the line of sight) passing through the q-th grid cell. By accurately calculating the truncation length of the part of the line of sight within each grid, a high-precision construction of the weight matrix is ​​achieved.

[0045] It's important to note here that columns in the W weight matrix that are all zeros need to be removed, meaning grids that are not visible to any line of sight. Similarly, rows in the W weight matrix that are all zeros need to be removed, meaning lines of sight that don't pass through the target region need to be deleted. This is done to ensure more accurate execution in subsequent iterations and to avoid NaN (Net-Analog Array) errors, where the matrix elements become infinitely large.

[0046] Step 2: Divide the target's imaging space into a grid, converting the continuous scene into discrete grid cells.

[0047] Based on the topological meshing of the "target geometry," for targets with special geometries such as tokamaks, the Cartesian coordinate system's block mesh is abandoned in favor of a curved mesh that fits the target contour. In ideal tokamak operation modes, or even considering only the ground state, plasma parameters (such as electron temperature and density) and neutral particle radiation distribution exhibit high axisymmetry. While the traditional Cartesian coordinate system (XYZ) mesh is universal, it cannot utilize this symmetry, and its flat boundaries introduce significant discretization errors when cutting through curved plasma (especially the D-shaped structure near the X point and the target plate), resulting in jagged edges in the reconstructed image. Specifically:

[0048] Step 1: 2D Regular Discretization;

[0049] On the polar section (RZ plane) of the device, a rectangular bounding box is defined for the imaging region. This region is divided into uniformly sized square pixel grid cells, with a grid side length of L = 1 cm. Each grid cell is defined by its center coordinates (R... i Z j Unique identifier. Compared to non-uniform meshes, regular meshes have a uniform spatial resolution, which facilitates subsequent error analysis and gradient calculation.

[0050] Step 2: 3D Ring Generation;

[0051] Utilizing the axisymmetric properties of plasma, each of the above square grids (R) i Z jIt extends circumferentially. Geometrically, this forms a toroidal ring with a square cross-section.

[0052] This invention employs a uniform square mesh, avoiding the distorted cells that may occur with non-uniform or curved meshes, ensuring a relatively uniform norm distribution of the weight matrix W, thereby improving the numerical stability of the inversion equation solution. A uniform step size of 1 cm ensures consistent spatial frequency response across the entire imaging region, avoiding resolution fluctuations caused by mesh size variations and facilitating quantitative analysis of the fineness of the radiation structure. It innovatively combines a "square mesh" with an "axisymmetric rotation" model, simplifying complex three-dimensional tomographic imaging problems (with unknowns on the order of 10^10^11). 5 Successfully reduced the dimensionality to a two-dimensional plane for solution (number of unknowns on the order of 10). 3 While significantly reducing computational costs, it utilizes the circumferential integral effect to improve the signal-to-noise ratio of the reconstruction results, making real-time reconstruction possible.

[0053] Step 3: Based on the grid division results and the line-of-sight information from the top-down view, construct a weight matrix that reflects the depth and density of the line of sight.

[0054] When constructing the weight matrix W, the intersection of the top-view line of sight with these "square-section rings" is calculated. Due to the top-view perspective, a straight line of sight will successively pass through multiple rings with different radii and heights. This step uses analytical geometry to calculate the chord length of the line of sight within each ring, which is used as a matrix element W. mn The value of . This means that the matrix element W mn Physically, it represents the radiation contribution of the nth annular body to the mth line of sight.

[0055] Subsequently, as Figure 3 As shown in Figure (b), constructing the weight matrix W includes analyzing the camera's line-of-sight depth from a top-down perspective and generating a camera line-of-sight depth map; then, by integrating the depth information, calculating the values ​​of each element of the weight matrix W. Because a top-down perspective is used, the line-of-sight density distribution is more uniform, which helps to increase the condition number of the matrix W, thereby improving reconstruction stability.

[0056] In existing tomographic imaging or 3D reconstruction techniques, calculating the intersection length between the line of sight (ray) and the mesh model is the core step in constructing the weight matrix. Traditional methods typically employ a global traversal strategy, that is, traversing every mesh cell in the model and calculating the geometric intercept relationship between each line of sight and that mesh.

[0057] Suppose the modeled three-dimensional space is divided into P cross-sections, each cross-section contains M grid cells, and the total number of sight lines is N. The time complexity of the traditional algorithm is as high as O(MNP). Taking the modeling of a typical tokamak device as an example, if a single cross-section contains 7,350 grids and is circumferentially divided into 360 planes, the total number of grids is as high as 2.646×10 6 to the power of 6. At this scale, global traversal not only causes a huge amount of computational redundancy, resulting in too high time costs, but also it is difficult to meet the rapid verification and iterative requirements of large-scale weight matrices, seriously restricting the operation efficiency.

[0058] To solve the above problems, this embodiment proposes a fast ray tracing method based on grid topological adjacency relationships. This method utilizes the characteristics of "sparsity" and "continuity" of the grids passed through by the sight line in the three-dimensional space, and significantly reduces the time complexity to the order of O(N) or O(PN). The specific implementation steps are as follows:

[0059] The first step: Determine the initial incident surface and incident point (initialization);

[0060] [[ID=ll]]First of all, it is necessary to determine the first grid position where the sight line enters the modeling area, as shown in Figure (a). Figure 4 According to prior knowledge, the position of the first surface grid passed through by the sight line when entering the modeling area can be known. Each surface is as shown in Figure (b).

[0061] Calculate the intersection points of the ray with the six surfaces of the first surface grid passed through. According to the spatial geometry theory, the number of intersection points of a ray with a convex polyhedron is usually 2 (passing through) or 0 (not passing through). Note: The situation where the sight line passes through an edge or a vertex is regarded as having an intersection point. Figure 4 As shown in Figure (b).

[0062] Calculate the intersection points of the ray with the six surfaces of the first surface grid passed through. According to the spatial geometry theory, the number of intersection points of a ray with a convex polyhedron is usually 2 (passing through) or 0 (not passing through). Note: The situation where the sight line passes through an edge or a vertex is regarded as having an intersection point.

[0063] If there are two intersection points A and B, calculate the Euclidean distances OA and OB between the two intersection points and the ray light source (or optical center) respectively.

[0064] Compare the distances: The point with the smaller distance is determined as the initial incident point, and the surface where it is located is the initial incident surface; the point with the larger distance is determined as the exit point.

[0065] For example, if OA < OB, then A is the incident point, and the first grid cell in the model can be directly indexed by the geometric surface where point A is located.

[0066] Determine the position index (i, j, k) of the incident point through mathematical knowledge (the i-th surface grid, the j-th row of this surface grid, the k-th column of this surface grid).

[0067] The second step: Path tracing based on adjacency relationships (iterative calculation); <000,0164>

[0068] After determining the incident point and position index of the current mesh cell, chain tracing is performed using the geometric properties of the mesh cell:

[0069] 1. Intercept Calculation: Based on the known parameters of the current mesh cell and the coordinates of the incident point, call the intercept length calculation module (e.g., execute the function "GetLineOverTrapezoidalCylinderLength") to calculate the coordinates of the exit point and the exit surface index of the ray in the current mesh.

[0070] 2. Obtaining line segment length: Based on the coordinates of the incident and exit points, calculate the length of the line segment intercepted by the current grid and store it in the weight matrix.

[0071] 3. Topology Index Update: Based on the principle of topological continuity that "the outgoing surface of the current grid is the incoming surface of the next adjacent grid", the spatial index of the next grid is directly derived.

[0072] Let the current grid index be (i, j, k) (the i-th grid, the j-th row of the grid, and the k-th column of the grid). If the calculated outgoing surface is the "top surface" of this grid, then the next grid must be located above this surface, and the index is updated to (i, j-1, k); similarly, if the outgoing surface is the bottom surface, the index is updated to (i, j+1, k).

[0073] 4. Loop Termination: Repeat the above steps until the derived next grid index exceeds the modeling range, at which point the algorithm ends. Calculate the chord length of the line segment within each grid for each line of sight, using it as the weight of the corresponding position in the weight matrix W. This results in an MM×N two-dimensional matrix, where MM is the number of lines of sight and N is the number of grids.

[0074] In real-world physical scenarios, the mesh that the line of sight passes through represents only a very small percentage of the total mesh in the model. For example, the aforementioned 2.646 × 10 6 Taking a single mesh model as an example, the actual number of meshes traversed by a single line of sight is typically only tens to hundreds, accounting for less than 1 / 10000 = 0.01%. This method avoids invalid calculations of irrelevant meshes, reducing the computational load by approximately four orders of magnitude. Even with existing techniques utilizing circumferential symmetry (simplifying the 3D problem to a 2D circular cylindrical section), traditional methods still require traversing all meshes of a single section (e.g., 7350 times). The method of this invention, however, eliminates the need for traversal, requiring only the calculation of a few dozen actual intersections, resulting in a computational efficiency improvement of nearly 100 times.

[0075] This invention significantly reduces the total number of floating-point operations, thereby reducing accumulated errors and improving the fault tolerance of weight matrix verification. Its high computational performance provides a powerful and practical tool for the real-time diagnosis and analysis of complex 3D structures. As shown in Table 1, compared to the previous brute-force solution, the path tracing based on adjacency relationships has a time cost of only 1 / 1000, while the space cost remains essentially the same, eliminating the time overhead of traversing meshes not visited by the line of sight.

[0076] Table 1

[0077]

[0078] Step 4: As Figure 5 As shown in Figure (a), the initial image vector S0 is calculated by combining the weight matrix with the input initial radiation distribution vector E0:

[0079] ;

[0080] Where: S0 is an M×1 dimensional vector, M is the total number of pixels in the image; W is an M×N dimensional weight matrix; E0 is an N×1 dimensional initial radiative distribution.

[0081] Step 5: As Figure 6 As shown in Figure (a), the initial matrix of all 10s is iteratively updated using the initial image vector S0, with the goal of minimizing the difference between the original observed signal and the reconstructed signal. After iteration, as shown in Figure (a), the initial matrix of all 10s is updated. Figure 6 Figure (b) shows the successful reconstruction of the image vector S from random noise, as shown. Figure 5 The reconstructed radiation distribution vector E is obtained from Figure (b).

[0082] This approach abandons the previous SART (Simultaneous Algebraic Reconstruction) technique and adopts Conjugate Gradient Least Squares (CGLS). The CGLS algorithm is an iterative process with feedback. The weight matrix is ​​known. and the initial synthesized image Solve the reconstruction objective .

[0083] A) Initialization phase:

[0084] Iterate starting from a full 10 vector: ;

[0085] Calculate the initial measurement residuals (The difference between the current predicted value and the actual measured value): ;

[0086] Calculate the initial projection residual (gradient). : ;

[0087] Set the initial conjugate search direction vector. : ;

[0088] B) Iterative Stage ):

[0089] In each iteration In this case, perform the following 6 standard calculation steps:

[0090] Step 5.1: Calculate the current step size factor (to ensure the error decreases the fastest in this search direction): ;

[0091] Step 5.2: Update the reconstruction results (stepping one step along the conjugate direction): ;

[0092] Step 5.3: Update measurement residuals: ;

[0093] Step 5.4: Update the projected residuals (normal equation residuals): ;

[0094] Step 5.5: Calculate the conjugate orthogonalization weighting coefficients: ;

[0095] Step 5.6: Construct the next conjugate search direction: ;

[0096] The iteration phase is repeated continuously until the preset maximum number of iterations, or the projected residual, is reached. Less than a certain extremely small tolerance threshold.

[0097] Algebraic reconstruction algorithms such as SART and SIRT (Simultaneous Iterative Reconstruction Technique) typically require hundreds or even thousands of iterations to converge. CGLS, however, utilizes conjugate orthogonal directions and usually requires only 20-50 iterations to achieve or even surpass the reconstruction accuracy of SART, significantly reducing computation time. As shown in Table 2, compared to SART, which requires more iterations and longer iteration times to achieve 1% MRE, CGLS takes only 1 / 125th the time of SART to achieve 1% MRE, demonstrating extremely fast convergence.

[0098] Table 2

[0099]

[0100] like Figure 7As shown, the MRE between the initial image vector S0 and the reconstructed image S is calculated. The results show that, thanks to the optimization of the viewpoint and the robustness of the algorithm, the MRE value of this method is as low as 0.00183514 (approximately 0.18%), achieving high-precision reconstruction under a wide field of view.

[0101] In use, this invention allows for the rapid acquisition of inverted radiation distribution simply by inputting images captured by a camera.

[0102] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0103] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0104] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0105] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0106] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0107] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

[0108] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for wide-field-of-view camera image reconstruction based on a top-down perspective, characterized in that, include: Step 1: Configure the camera from a tangential viewpoint to a top-down viewpoint. By adjusting the position of the optical center and the pitch and azimuth angles of the direction vector, construct the camera imaging geometry model from the top-down viewpoint to obtain full and partial line-of-sight information. Step 2: Divide the target's imaging space into a grid, converting the continuous scene into discrete grid units; Step 3: Based on the grid division results and the line-of-sight information from the top-down view, construct a weight matrix that reflects the depth and density of the line of sight; Step 4: Combine the weight matrix with the input initial radiation distribution vector to generate the initial image vector; Step 5: Based on the initial image vector, with the goal of minimizing the difference between the original observed signal and the reconstructed signal, the radiation distribution is continuously updated through an iterative algorithm until the reconstruction error converges, and the final reconstruction result is output.

2. The method for wide-field-of-view camera image reconstruction based on a top-down perspective according to claim 1, characterized in that, Step 1, obtaining partial line-of-sight information, includes: calculating the intersection of the full line-of-sight with the boundary of the target area, truncating the full line-of-sight, and obtaining line segments located only within the target area as partial lines of sight; if a certain full line-of-sight does not pass through the target area, the pixel corresponding to the line-of-sight is marked as invalid in the reconstruction calculation, and its corresponding row in the weight matrix is ​​all zero.

3. The method for wide-field-of-view camera image reconstruction based on a top-down perspective according to claim 1, characterized in that, Step 2, which involves dividing the imaging space into grids, includes: defining a rectangular bounding box on the poloidal section of the target, dividing it into square pixel grids of uniform size, and utilizing the axial symmetry of the plasma to extend each square grid around the circumference, forming a ring with a square cross-section.

4. The method for wide-field-of-view camera image reconstruction based on a top-down perspective according to claim 1, characterized in that, The construction of the weight matrix in step 3 includes: using a fast ray tracing method based on grid topology adjacency, first determining the initial incident surface and incident point of the line of sight entering the modeling region; based on the topological continuity that the exit surface of the current grid is the incident surface of the next adjacent grid, deriving the spatial index of the next grid; calculating the chord length of the line of sight inside each grid as the weight, and obtaining the weight matrix.

5. The method for wide-field-of-view camera image reconstruction based on a top-down perspective according to claim 4, characterized in that, The steps for determining the initial incident surface and incident point include: calculating the intersection points of the ray with the six surfaces of the first grid; if there are two intersection points, calculating the Euclidean distance between the two intersection points and the ray source; the point with the smaller distance is determined as the initial incident point, and the surface on which it is located is the initial incident surface; the point with the larger distance is determined as the exit point.

6. The method for wide-field-of-view camera image reconstruction based on a top-down perspective according to claim 1, characterized in that, After constructing the weight matrix in step 3, the method further includes: deleting columns in the weight matrix where all columns are zero, so as to delete grid cells that are not visible to all lines of sight; and deleting rows in the weight matrix where all rows are zero, so as to delete lines of sight that do not pass through the target area.

7. The method for wide-field-of-view camera image reconstruction based on a top-down perspective according to claim 1, characterized in that, The iterative algorithm in step 5 employs the conjugate gradient least squares method, including: Initialization phase: Set the initial radiation distribution vector, calculate the initial measurement residual based on the weight matrix and the initial radiation distribution vector, calculate the initial projection residual based on the transpose of the weight matrix and the initial measurement residual, and set the initial conjugate search direction vector to be equal to the initial projection residual; Iteration phase: In each iteration, the step size factor is calculated based on the relationship between the current projection residual, the weight matrix, and the current search direction vector. The reconstruction result, measurement residual, and projection residual are updated sequentially using this step size factor. The conjugate orthogonalization weight coefficient is calculated based on the norm ratio of the projection residual before and after the update. Then, the updated projection residual is combined with the current search direction vector based on this weight coefficient to construct the next conjugate search direction. The above iteration phase is repeated until the preset iteration termination condition is reached.

8. The method for wide-field-of-view camera image reconstruction based on a top-down perspective according to claim 1, characterized in that, The method for determining the convergence of reconstruction error in step 5 is as follows: calculate the average relative error between the initial image vector and the reconstructed image vector, and determine convergence when the average relative error meets a preset threshold.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When one or more programs are executed by the one or more processors, the one or more processors implement the wide-field-of-view camera image reconstruction method according to any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, It stores executable instructions that, when executed by a processor, enable the processor to implement the wide-field camera image reconstruction method based on a top-down view as described in any one of claims 1-8.

Images